Technology / Core AI
Cost-effective AI detector development, built on computer vision expertise, reusable models and in-house Active ML infrastructure. Focus development effort on adapting detectors to your real-world processes.
Expertise backed by evaluation
3DiVi Face SDK takes part in NIST FRTE 1:1 verification through a technology partner. This provides an external evaluation context for our face-recognition technology.
For a custom computer vision project, the essential questions are equally concrete: what must be detected, on which data, under which conditions, and against which acceptance criteria?
Explore face recognition benchmarks ↗From algorithms to applied projects
Explore recorded transport demonstrations and a published manufacturing pilot. Each example connects camera footage to a defined event or operational question.
Detect boarding and alighting, driver distraction and pedestrians in camera-covered blind spots. Assess the camera setup, target hardware and integration requirements for your transport solution.
Comparative test reports can be discussed with our team. The deployment story below describes a tram project in Russia.
Connect hand and body tracking, work-zone interactions and object detection to questions about process sequences, equipment downtime and tool use.
In the published factory pilot, two hours of operation revealed 25 instances of technological downtime and 8 unjustified downtimes. These results describe that pilot.
Active ML / in-house infrastructure
Core AI brings together reusable models, a deep-learning development platform and compute infrastructure. Active ML is our in-house infrastructure for developing and improving detectors around project requirements.
Reuse computer vision building blocks for face recognition, body tracking, objects and actions. Focus development effort on the data and events specific to your task.
Define the camera views, examples and labels for your task. Train or adapt models and inspect the errors that matter to the application.
Assess model quality alongside latency, computing resources and integration requirements. Agree the target configuration and acceptance criteria for the project.
Learn more about the Core AI technology foundation on our company page ↗.
Project workflow
Describe the object, action, sequence or visual condition to recognize. Define what the receiving system should do with the result.
Collect representative examples and agree labels, metrics and a separate test dataset for the intended operating conditions.
Train or adapt the detector. Review false positives, missed events and difficult examples, then refine the approach.
Measure quality on the evaluation data. Check the camera setup and performance requirements of the target platform.
Define the event outputs and software integration. Use Omni Agent where its capabilities fit the deployment.
Use observed error cases and changes in conditions to scope the next validation and adaptation cycle.
How the technologies fit together
Computer vision expertise, models and infrastructure behind project-specific detector development.
Explore the foundation ↓In-house infrastructure supporting the iterative work of developing and improving AI detectors.
Discuss your requirements ↓Video-stream processing, detection scenarios and event integration for products and operational systems.
Explore Omni Agent ↗Questions about the technology
Core AI is the technology foundation behind 3DiVi computer vision development: reusable models, a deep-learning development platform and compute infrastructure. It supports our work on project-specific models and detectors.
Core AI describes the expertise and development foundation. Omni Agent is video analytics software for processing video and integrating detections into an application. Start with Omni Agent or a project discussion when you need a deployable solution.
Active ML is 3DiVi's in-house infrastructure for AI detector development and improvement. The data workflow, model adaptation, evaluation process and delivery scope are defined for the particular project.
No. NIST FRTE evaluates face-recognition algorithms. APC, driver monitoring, blind-spot detection and process-monitoring scenarios need their own datasets, metrics and acceptance tests.
Yes. Share the objects, defects or missing components you want to detect, together with representative images and the camera setup. Feasibility, data needs and a validation pilot must be assessed for that inspection task.
Describe the event to detect, the intended process, available footage, camera placement, target hardware and integration requirements. Agree the quality metrics and operating conditions before a pilot.
Tell us what your system needs to detect or verify. Discuss the data, evaluation criteria and integration path with the 3DiVi team.
